SocialGaze: Improving the Integration of Human Social Norms in Large Language Models

TL;DR

This paper introduces SocialGaze, a multi-step prompting framework that enhances large language models' ability to judge social acceptance, aligning their judgments more closely with human social norms and revealing biases related to gender and age.

Contribution

The paper presents SocialGaze, a novel multi-step prompting method that improves LLMs' understanding of social norms and reduces biases in social judgment tasks.

Findings

SocialGaze improves alignment with human judgments by up to 11 F1 points.

LLMs show gender bias, judging males more unfairly.

Older narrators' social judgments are more aligned with humans.

Abstract

While much research has explored enhancing the reasoning capabilities of large language models (LLMs) in the last few years, there is a gap in understanding the alignment of these models with social values and norms. We introduce the task of judging social acceptance. Social acceptance requires models to judge and rationalize the acceptability of people's actions in social situations. For example, is it socially acceptable for a neighbor to ask others in the community to keep their pets indoors at night? We find that LLMs' understanding of social acceptance is often misaligned with human consensus. To alleviate this, we introduce SocialGaze, a multi-step prompting framework, in which a language model verbalizes a social situation from multiple perspectives before forming a judgment. Our experiments demonstrate that the SocialGaze approach improves the alignment with human judgments by up to 11 F1 points with the GPT-3.5 model. We also identify biases and correlations in LLMs in assigning blame that is related to features such as the gender (males are significantly more likely to be judged unfairly) and age (LLMs are more aligned with humans for older narrators).